ServiceNow

When your laptop breaks at work, you open a ticket. When a new employee needs a desk, a login and a payslip, someone fills in a form that sets twenty people in motion. When a bank's system goes down at 3 a.m., an engineer is paged, the incident is logged, and the change that fixes it is approved by someone else. In most of the world's largest organisations, all of that runs on ServiceNow. It began as the help desk for IT departments and has become the system that routes work through a company — IT, human resources, customer service, security — sold by annual subscription. A ten-year-old's version: ServiceNow is the system where a big company writes down every request and makes sure somebody does it.

It is a superb business to own a share of. Some 97% of revenue is subscriptions; customers renew 98% of their contract value every year and have done so for as long as the company has reported it; revenue grew 24% in the second quarter of 2026 to $3.99bn; the company generates about 35 cents of free cash flow from each dollar of sales. An investor who bought at the 2012 IPO has made 38 times his money.

And yet at $137.76 on 1 October 2026 the shares are 43% below their January 2025 high, after falling as low as $81.24 in April. The market's fear has a name — the "SaaSpocalypse" — and a thesis: that AI agents will do the work of the people ServiceNow charges by the seat, and that the company is buying growth (three acquisitions for $11.7bn in nine months) because it cannot make enough of its own. This report asks how much of that fear is true, and whether the price is now fair for a business of this quality.

Two lessons for an owner. First, the customers never wobbled while the shareholders did. Through a fall of two-thirds, the renewal rate stayed at 98% and contracted revenue kept growing above 20%. Whatever AI eventually does to ServiceNow, it had not done it by the time the market priced it in. Second, the company answered the fear by spending: $11.7bn on acquisitions in nine months, by far the biggest buying spree in its history, at a time when its own shares were falling. Whether that was courage or haste is the capital-allocation question of Part VIII.

Two features of the revenue that an owner should know. First, government. One US federal channel partner accounted for 13% of revenue in the first half of 2026, up from 11%, and federal contracts cluster in the third quarter. In the second quarter federal demand pulled some on-premise revenue forward from the third, which flattered Q2 and will flatter the comparison against Q3. In the first quarter, delayed on-premise deals in the Middle East, during the war there, cost about 0.75 points of growth.

Second, the margin is moving the wrong way for a reason. The non-GAAP subscription gross margin was 80.5% in Q2, down from 83% a year earlier, because more customers run ServiceNow on hyperscalers' clouds and because AI work consumes paid computing. The new revenue is real; it is less profitable per dollar than the old. That is the first measurable sign of what AI does to a software company's economics, and it is worth watching more closely than any headline growth rate.

Buffett's 1999 letter explained why he owned no technology companies: "we have no insights into which participants in the tech field possess a truly durable competitive advantage." ServiceNow is as good an answer to that problem as software offers. A renewal rate of 98% for years on end, in a product customers configure for a decade, is a moat measured in the customers' own decisions. We score it 8: wide and stable at the core of the Global 2000, narrower in the mid-market, and facing a technology that could — not yet does — lower the cost of leaving. It is not 9 because, unlike Visa's or S&P Global's, the threat to it is real enough to have halved the share price.

Management's answer is unambiguous. "Are we worried about seat compression? Not at all," Bill McDermott said on the July call; customers who try to build their own agents find it costs "5 to 10x" more than running them on ServiceNow, and the theory that AI replaces software is "a parlor trick" (Fortune, 23 April 2026 — the day the shares fell 17.7%). He also told CNBC that active seats were up 25% — a figure that appears in no filing.

Our reading is narrower than either side. The seat is not dying at ServiceNow yet; nothing in the renewal rate or the backlog says it is. But the seat's share of the growth is shrinking, and ServiceNow is turning itself from a toll per person into a toll per unit of work. That can be a good business — arguably a better one, if AI multiplies the work done through the platform. Its first measurable cost is margin: the AI revenue carries computing costs the seat never did. The question an owner should watch is not whether revenue grows — it will — but whether the subscription margin stabilises above 80% while it does.

The pattern. No established competitor is winning ServiceNow's core customers; the evidence is a $42,000 account and a 98% renewal rate. The competition that matters is of a new kind — software that writes the software, agents that work across systems, start-ups that price by result — and ServiceNow's response so far has been to partner with the largest of those players and to buy the most promising small one. That is a sensible defence. It is also an expensive one, and it is the reason the balance sheet in Part VIII looks different from a year ago.

The buying spree. In nine months ServiceNow agreed to pay about $11.7bn for four companies, the largest being Armis, a cyber-security firm valued at $6.1bn in a private round in November 2025 and bought for $7.75bn a month later — about 23 times its $340m of annual recurring revenue. To pay for it the company borrowed $4bn in bonds and $2.1bn of commercial paper; a balance sheet with some $8.6bn of net cash at the end of 2025 now has roughly $0.8bn of net debt. Armis adds about 1.25 points of growth this year and takes about 2 points off the free-cash-flow margin.

The buybacks. In January the board added $5bn of repurchases and launched a $2bn accelerated buyback, "with a primary objective of managing the impact of dilution". In the same half-year the company granted 31.3m restricted shares — more than it bought back. Buffett's 2011 letter is blunt about this: "It doesn't suffice to say that repurchases are being made to offset the dilution from stock issuances."

Our verdict on management: an excellent operator, a less disciplined allocator. The product, the sales machine and the renewal rate are as good as any in software, and the promise to cut stock pay below 10% of revenue by 2029 is the right one. But a 37% pay rise in a down year, $11.7bn of acquisitions, the largest at a premium to a private valuation set weeks earlier, and buybacks that merely chase grants are not the record of an owner-minded allocator. We score management and capital 6.

The one subtraction that matters. ServiceNow reports an "adjusted" operating margin of about 30% and a GAAP margin of 4% in the same quarter. Most of the difference is stock-based pay — 16.4 points of revenue — with the amortisation of the companies it bought, deal costs and severance making up the rest. Buffett's 2015 letter called stock-based compensation "the most egregious example" of costs left out of adjusted earnings: "The very name says it all: 'compensation.' If compensation isn't an expense, what is it?" The amortisation is a fair adjustment — it is the accounting echo of acquisitions already paid for. The stock pay is not. We count it.

So the honest cash measure is free cash flow after stock pay: about 35% minus about 15% of revenue, or roughly 20 cents of every revenue dollar for 2026. That is still a very good business. It is not the 35-cent business the headline suggests.

What is it worth? We discounted owner earnings at 10% for ten years under three stories, starting from 2026 revenue of about $16.2bn and owner earnings of 20% of revenue. If AI shrinks the seat model — growth of 12% then 6%, margins stuck at 20%, 15 times at the end — about $74. Our conservative case — 15% then 10%, owner margin rising to 25% as stock pay falls as promised, 20 times — about $142. Management's own plan — roughly $30bn of subscription revenue in 2030, then 12% growth, a 28% owner margin and 22 times — about $205.

At $137.76 the price sits almost exactly on our conservative case. For a business with a 98% renewal rate, that is a fair price — not a bargain, and not a bubble. Our rule for businesses of the highest quality is to buy at a fair price rather than wait for a bargain; ServiceNow's moat qualifies, but its capital allocation and its exposure to an untested technology shift do not quite, which is why we would start small and add below about $115, some 20% under our conservative value. The shares were at $81 in April.